The Pink Tax: Why Do Women Pay More?
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
The Pink Tax:
Why Do Women Pay More?∗
Please click here for the most recent version.
Kayleigh Barnes† Jakob Brounstein‡
(Job Market Paper)
January 1, 2023
Abstract
We study the question of whether women, on average, pay a price premium — a
so-called “pink tax” — for the products they buy. A particular concern facing policy
makers is whether such differences are a form of gender based price discrimination.
Using scanner data, we find that averaged across the entire retail grocery consumption
basket, women pay 4% more per unit for goods in the same product-by-location market
as do men. This price differential is generated by a 15% higher average per unit price
paid by women on explicitly gendered products, like personal care items, as well as
a 3.8% higher average per unit price paid by women on ungendered products, like
packaged food items. Higher prices paid by women could be the result of differences in
demand elasticity, competitive structure, or sorting into goods with differing marginal
costs. To disentangle these mechanisms, we estimate demand differences between men
and women and structurally decompose price differences into markups and marginal
costs. We find that women are, on average, more price elastic consumers than men,
suggesting that as a consumer base women are not likely to be charged higher markups
under price discrimination. Overall, we find that the pink tax is not sustained by higher
markups charged to women, but by women sorting into goods with higher marginal costs
and lower markups.
∗
We would like to thank Ben Handel, Jon Kolstad, Ben Faber and Nano Barahona for their mentorship and
guidance. We thank Matt Backus, Mathilde Bombardini, Nina Buchmann, Bryan Chu, Giovanni Compiani,
Stefano DellaVigna, Thibault Fally, Jonathan Holmes, Leticia Juarez, Kei Kawai, Sarah Moshary, Mathilde
Muñoz, Nina Rousille, and Damian Vergara, as well as participants in the UC Berkeley IO and Trade lunch
seminars and University of Washington PHEnOM seminar for helpful comments. Researcher(s)’ own analyses
calculated (or derived) based in part on data from Nielsen Consumer LLC and marketing databases provided
through the NielsenIQ Datasets at the Kilts Center for Marketing Data Center at The University of Chicago
Booth School of Business. The conclusions drawn from the NielsenIQ data are those of the researcher(s) and
do not reflect the views of NielsenIQ. NielsenIQ is not responsible for, had no role in, and was not involved
in analyzing and preparing the results reported herein. While the categories of “men” and “women” do not
cover the entirety of the gender spectrum, our work focuses on these two categories. Unfortunately, data
indicating non-binary and alternate gender identities is scarce, and for the purposes of studying the Pink
Tax, producers and retailers have historically marketed goods and services toward either men or women.
†
University of California, Berkeley email: kayleighnb@berkeley.edu
‡
University of California, Berkeley email: jakob.brounstein@berkeley.edu1 Introduction
Is it more expensive to be a woman? Economic and societal forces have shaped preferences
and product offerings to create disparities in the way men and women consume goods. The
notion that there exists a price premium on women’s consumer goods relative to those of
men is colloquially referred to as the “pink tax”. The concept has received considerable
attention in popular media and has spurred recent legislation in New York and California.
This public discourse on the pink tax often attributes it to gender based price discrimination,
where goods that are marketed to women have higher markups resulting from less elastic
demand or less competitive markets. Existing studies of the pink tax find mixed evidence of
its scope and magnitude and either focus on a narrow set of goods or do not delve into its
underlying economic mechanisms (Moshary, Tuchman, and Bhatia 2021; Guittar et al. 2022;
NYCDCA 2015; Duesterhaus et al. 2011; Manzano-Antón, Martinez-Navarro, and Gavilan-
Bouzas 2018). Moshary, Tuchman, and Bhatia (2021) evaluate the existence of the pink tax
for personal care items and find no evidence of higher markups on women’s products when
controlling for proxies of marginal costs. Controlling for marginal costs restricts comparisons
to between goods with the same inputs and tests for third degree price discrimination. How-
ever, this type of comparison abstracts away from men’s and women’s purchase decisions and
does not capture the role of differential sorting by men and women.
This paper explores the existence and underlying mechanisms of the pink tax by describing
consumption baskets for men and women, analyzing how they vary by quantity, price, and di-
versity of products consumed, and then decomposing observed price differences into markups
and marginal costs. Our paper considers a broad definition of the pink tax1 , considering
any channel through which women may face higher markups in the retail consumer packaged
goods (CPG) space. This definition allows us to capture the role of differential sorting be-
tween men and women and second degree price discrimination, or versioning, in generating
the pink tax. We find that, averaged across the entire grocery consumption basket, women
pay 4% higher per unit prices than men do for products in the same product-by-location
market. We find that this price difference is sustained not just by purchases of gendered
1products, like men’s and women’s razors, but also by differences in purchasing habits be-
tween men and women for food and household items. This finding could be driven by three
economic mechanisms that determine pricing: (i) women could have less elastic demand than
men, (ii) women could consume products with more market power or from less competitive
markets than men, or (iii) women could consume products with higher marginal costs. Pric-
ing disparities due to markups based on demand differences or competitive structure impact
consumer surplus directly, potentially driving welfare differences by gender. Price differences
based on underlying production costs across the consumption baskets, on the other hand, do
not reduce consumer surplus and are not perceived as an issue for “fairness” (Kahneman,
Knetsch, and Thaler 1986). Disentangling the mechanisms driving the observed price pre-
mium on women’s products is, thus, important to inform economic understanding and policy
alternatives.
To characterize the Pink Tax and, broadly, gender differences in consumption habits, we
employ several data sets that contain detailed information on individuals and their pur-
chases, store-level product offerings, and retail prices. The Nielsen Consumer Panel Survey
features a 15-year rotating panel of households and the near-universe of their purchases at
big box retailers and grocery stores. Importantly, the data includes rich household demo-
graphic information as well as highly detailed product and purchase characteristics—including
deal/sale usage, prices paid/quantities consumed, and a hierarchy that aggregates products
into tractable market definitions. By restricting the bulk of our analysis to single-member
households, we are able to attribute each purchase made to a specific gender. We augment
the Consumer Panel with the Nielsen Retailer Scanner data which contains store level data
on prices and quantities sold in any given week.
1
We identify three scenarios through which the pink tax may operate: 1) different prices for goods with
the identical inputs: e.g. without changing anything else, by coloring a product pink, retailers and producers
can charge a higher price. 2) different prices for goods with identical uses but non-identical inputs: i.e.
the price difference between goods purchased by men or women is attributable to differences in the cost of
production. 3) expense differences driven by goods that are almost exclusively purchased by a single gender:
e.g. the purchase of makeup or feminine hygiene products. In some instances, the pink tax refers to the
luxury, sales, or value added taxes statutorily placed on women’s hygienic products. Our analysis focuses on
the more general case of price differences between men’s and women’s consumer goods.
2We begin by establishing the existence of systematic gender differences in consumption and
pricing along two margins: consumer behavior and the product space. To document consumer
behavior, we describe consumption bundles for men and women, documenting differences in
their unit price and composition. We find that women spend about 6% more than men do
on retail CPG consumption and that their consumption bundles are larger and more diverse.
The products that women purchase are on average 4% more expensive per unit than those
purchased by men in the same product-by-location market. In the product space, we doc-
ument a significant share of products that are exclusively bought by one gender, with the
majority of these products gendered towards women. These products are particularly com-
mon in markets with explicit gender differentiation in marketing and product design, such as
in beauty and personal care goods. We categorize products bought at least 90% of the time
by one gender as “gendered” products, categorizing all other products as “ungendered”. We
then decompose the average 4% price premium paid by women into a contribution from differ-
ential sorting into ungendered products and from purchases of explicitly gendered products,
finding that women pay an average of 3.9% higher prices on ungendered products relative to
men and that women pay an average of 15% higher prices on gendered products relative to
men. While gendered items have large price premiums, they make up a small share of actual
purchases; the bulk of the price premium is being driven by women buying more expensive
ungendered items than men.
We then turn our attention to understanding the demand and supply mechanisms that give
rise to women paying higher prices. Profit maximizing firms set prices as a function of own-
price elasticities, market shares, cross-price elasticities of products owned by the same parent
company, and marginal costs. Less elastic own-price elasticities put upward pressure on prices
as firms can raise prices without losing much of their consumer base. Higher prices paid by
women could then be consistent with women being less elastic and firms price discriminating
off of the gender composition of their consumer base. Alternatively, differences in compe-
tition and market structure can also contribute to higher markups if women’s markets are
more concentrated, meaning that their products have higher market shares, or if women’s
products are more likely to be owned by multi-product firms, as substitution to products
3with the same parent company puts upward pressure on prices. Both the demand elasticity
and competition narratives would contribute to higher prices through women paying higher
markups, which has potential welfare effects for women. Finally, women could face higher
prices if the products that they prefer have higher marginal costs than the products that men
prefer, that is, if women differentially sort into products with higher costs of production.
To assess these possibilities, we model demand and supply, attributing differences in pric-
ing and product choice to markups and marginal costs. We begin by estimating demand
elasticity differences between men and women across the entire retail grocery consumption
basket. We develop a simple, tractable model assuming constant elasticity of substitution
that allows us to estimate demand by gender in the aggregate population. We aggregate
individual-level purchase data to the a gender-by-product module-by-location market level
and we find that, on average, women consume products more elastically than do men. This
finding is consistent with women being the consumer group that is charged lower markups
rather than higher markups under price discrimination.
We have shown that, on average, women are more elastic consumers than men, but in order to
better understand product specific demand and decompose prices into markups and marginal
costs we estimate a differentiated products demand and supply model. This model incorpo-
rates market structure and allows for flexible substitution patterns based on how “gendered”
a product is. We focus on five markets: yogurt, protein bars, disposable razors, deodorant
and shampoo. We selected yogurt because its pricing patterns mirror the descriptive analysis
of the full consumption basket, it is representative of the most commonly bought item in the
data, a packaged food item, and it has a moderate amount of differential sorting by gender.
The other four markets were selected because gender is an explicit component of marketing
and product design, allowing for identification of demand for explicitly gendered products
which might not be well captured in the CES model.
To estimate the model, we use store-level data on quantities and prices. With this data,
we gain improved inference on markets where purchase frequencies make individual-level
4data sparse, like personal care items. We allow for heterogeneity in preferences for the gen-
der of a product and instrument for prices with Hausman instruments and retail chain-level
leave-out mean prices following evidence from DellaVigna and Gentzkow (2019). To map our
results back to consumer demographics, we analyze how our results vary with the product’s
woman purchase share observed in the individual-level purchase data. We find that women’s
products are either more elastic or have no significant differences in elasticities from ungen-
dered or men’s products and that women’s products have lower markups and higher marginal
costs. These results, while allowing for more granular identification and flexible demand, are
largely consistent with the CES demand analysis. Overall, our findings imply that observed
price differences between men and women are primarily driven by women sorting into higher
cost products.
Our findings suggest that the pink tax is not a form of systemic price discrimination against
women but that, if anything, women pay lower markups on average than men. Current leg-
islation is largely focused on banning price differences for products that differ only in gender.
Our paper suggests that these laws are likely to be ineffective at addressing price disparities
between men and women, as the majority of our observed pink tax can be explained by
men and women sorting into products that differ by more than just gender.2 Our findings
have important implications for other policy relevant issues, like potential disparities in the
incidence of inflation between men and women. Finally, our findings motivate future re-
search to study how men’s and women’s preference differences are formed as well as the role
of preference differences in generating product differentiation through product entry and exit.
We proceed as follows: Section 2 discusses the background and history of the pink tax
as well as relevant literature. Section 3 describes our data. Section 4 presents our descrip-
tive analysis. Section 5 describes and estimates a constant elasticity of substitution demand
model of men and women’s consumption. Section 6 describes and estimates a differentiated
2
The state of New York has banned pricing on the basis of gender through bill S2679 which took effect
in 2020. A similar bill, AB 1287, was signed into law in California by Governor Gavin Newsom on Sept. 27,
2022. The Pink Tax Repeal Act has been presented in Congress four times and aims to put national law in
place similar to the New York and California policy.
5products demand model. Section 7 discusses the implications of our results and concludes.
2 Background and Literature Review
The term “The Pink Tax” was first coined in the 1990’s in California, when concerns about
gendered price discrimination of services, such as dry cleaning and in hair salons led to the
explicit anti-price discrimination law, The Gender Tax Repeal Act. Soon after similar leg-
islation was passed in New York City and Miami. A national version of The Gender Tax
Repeal Act has been introduced federally several times since 2016 but has never been passed.
More recently, there has been renewed policy interest in the Pink Tax, particularly in the
setting of gendered price discrimination for consumer retail products. In 2020, the state of
New York passed legislation that would outlaw gender differential pricing. In 2022, California
passed a similar law. The language surrounding these laws frames the Pink Tax as a price
discrimination story with the underlying assumption that markups are higher for women’s
products.
However, in spite of its importance as a potential component of gender inequality and its
wide presence in popular discussion, there are few studies that rigorously substantiate the
Pink Tax. The New York law was based on evidence collected and presented in a New York
Department of Consumer Affairs (NYC DCA) study in 2015. The NYC DCA compares
products in thirty-five categories and five broader industries with “clear male and female ver-
sions” sold by New York City retailers, finding that women’s products cost on average seven
percent more than similar products for men. While it provides key preliminary suggestive
evidence of a “pink tax”, the NYC DCA analysis is largely incomplete: it consists of a highly
limited number of goods that were gender-matched in a subjective manner; moreover, it
only documents raw list price differences rather than actual prices paid. Recently, Moshary,
Tuchman, and Bhatia (2021) assess the pink tax under the definition in the New York law,
products that differ only in gender. They control for brands and ingredients as a proxy for
marginal costs and find no evidence of a systemic pink tax. Other works similarly focus on
health and beauty products, using in store surveys of products to descriptively document
6price premia of around 5% on women’s goods. (Duesterhaus et al. 2011; Manzano-Antón,
Martinez-Navarro, and Gavilan-Bouzas 2018; Manatis-Lornell et al. 2019) Taken together,
list price differences suggest that women may be paying more than men for goods with similar
uses, but Moshary, Tuchman, and Bhatia (2021)’s finding of no pink tax when matching on
marginal costs suggests that women and men are sorting into products that differ by more
than just gender. Our paper explicitly studies differences in the prices, markups and marginal
costs of the entire range of retail goods that are bought by men and women, capturing this
sorting component.
Within economics, there is relatively little work that focuses on gender disparities in the
pricing of goods and services. The most-related work on gendered price discrimination fo-
cuses on bargaining contexts for wages or products. Most recently, Rousille (2021) attributes
nearly 100% of gender pay inequality among tech industry workers to differences in wage-
asks by interviewees, underscoring the potential role for differences in bargaining power to
generate gender-inequality. Ayres and Siegelman (1995) provide evidence of race and gender
discrimination in bargaining for new cars, finding that women and black men paid signifi-
cantly higher markups for cars than white men. This setting has been further studied by
Goldberg 1996 and Trégouët 2015, with Castillo et al. (2013) also documenting systematic
differences in stages of taxi-price bargaining for men and women. Fitzpatrick 2017 finds
evidence of gender price discrimination in the context of bargaining for anti-malarial drugs.
While these studies provide evidence and precedence of price discrimination against women,
they do not capture a mechanism by which price discrimination can occur of goods with
simple take-it-or-leave-it list prices nor the role of differences in preferences across product
offerings.
Because we investigate the pink tax across the retail consumption basket, we view this work
as closely-related to research on inequality in consumption and product offerings. Jaravel
2019 finds that poorer households experience higher inflation and price indices, exacerbating
income inequality in real terms. Though we do not directly calculate differences in inflation
for men and women, our work on gender explores a new angle through which price index
7inequality may shape wealth inequality at large. Aguiar and Hurst 2005 use survey data to
demonstrate that consumption remains relatively constant among individuals as they transi-
tion into retirement, simultaneously documenting differences in sources of consumption (e.g.
restaurant dining, home-production, etc.) between men and women. Aguiar and Hurst 2007
also quantify objects such as the substitution elasticity between shopping and home produc-
tion and and the willingness to engaging price shopping or to take advantage of deal; while
not explicitly focused on gender distinctions, their findings on the price returns of time spent
shopping have important implications for understanding the differences in prices paid by men
and women.
The implications of the Pink Tax for gender equality are wide-reaching: taking into account
differences in the cost of consumption prompts us to re-frame the widely-studied difference in
wages between men and women as a nominal wage gap. Moretti (2013) has shown that pop-
ulation specific price indices have important implications for wage inequality in real terms.
Estimates of the raw gender pay gap tend to around 20% today, decreasing to about 10%
after including differences in qualifications (Blau and Kahn (2017)). The presence of an ag-
gregate Pink Tax on women’s consumption augments these inequalities by reducing women’s
purchasing power. Moreover, by accounting principles, the existence of a Pink Tax also
highlights differences in overall consumption and savings between men and women. Women,
facing on average higher prices for their respective consumption bundles face both lower real
wages and potentially lower scope to accrue lifetime savings and consume.
Faber and Fally 2022 study how product offerings and firm sorting drive price index in-
equality across incomes. They find that larger, more productive firms endogenously sort
into markets that cater to richer households and that this drives asymmetry in price indices
across the income distribution. This study suggests that supply side factors may play an
important role in differences in product offerings and marginal costs for men and women.
Simultaneously, DellaVigna and Gentzkow 2019 find substantial price mis-optimization for
retail chains, where stores typically implementing uniform prices throughout all US stores
irrespective of local demand and cost factors—suggesting some limitations to how and to
8what extent firms engage in optimal price strategy. B. J. Bronnenberg et al. 2015 study how
information and experience may drive inequality in product choice on the consumer side by
looking at differences in choices made between experts (by profession) and non-experts in
purchasing drugs and grocery items. They find that non-experts over pay for brand name
products more than experts do. While expertise may not be a direct driver of differences in
product choices between genders, this work highlights the potential for misinformation and
incorrect product beliefs to affect choices and prices paid.
We contribute to the large literature in industrial organization on the role of product differ-
entiation within markets. Berry, Levinsohn, and Pakes (1995) developed the method that
we use to estimate a discrete choice model in the presence of product differentiation. We
incorporate the gendered-ness of a product as a characteristic over which individuals may
have heterogeneous tastes. We use a revealed preference approach to identifying product
gender, which means we do not need existing product characteristic data which enables us to
estimate demand in multiple markets. Economists have long thought about the role of het-
erogeneous tastes and product differentiation in welfare. (George J Stigler and Becker 1977;
Spence 1976) Product differentiation and price discrimination are sometimes thought of as
separate phenomenons but in a broad view of price discrimination as any markup difference
between consumers groups (like that of George J. Stigler 1987) the two are linked. Shapiro
(1982) discusses second degree price discrimination through versioning, but there is no clear
line of where versioning ends and product differentiation begins. Our paper contributes to
this literature by analyzing how demand composition for differentiated products may lead to
higher markups placed on a particular demographic group.
Though we do not estimate a general equilibrium model that incorporates endogenous prod-
uct entry and exit, our findings suggest a natural next step of examining firms decisions to
produce gendered products. Wollmann (2018) models entry and exit decisions of truck mod-
els finding that allowing for entry and exit moderates price increases resulting from mergers.
Barahona et al. 2020 finds that firms decision to reformulate after a policy that affects demand
depends on expected profits that face a tradeoff between bunching product characteristics
9that appeal to a larger demand group but face higher competition or differentiating a product
to a smaller consumer base but facing less competition. Firm’s incentives to innovate and
introduce new products have also been studied in the trade literature. Work on firm and
product heterogeneity stemming from differences in demand and costs among firms finds im-
plications for innovation and competition. ( Hottman, Redding, and Weinstein 2016, Faber
and Fally 2022, and Atkeson and Burstein 2008) Broda and Weinstein (2010) and Bernard,
Redding, and Schott (2010) also substantiate the role of innovation and turnover in driving
evolution in prices, finding a substantial amount of product innovation: namely that half of
firms switch their products within a span of five years and that product creation is a much
stronger component of net product entry than product destruction. Although these works
do not focus on gender or gendered product spaces explicitly, their findings have important
implications for how we understand and motivate study of men’s and women’s consumer
goods markets.
3 Data
We combine data on from two main sources and two supplemental sources to conduct our
analysis. Our main analyses rely on the NielsenIQ data including the HomeScan Panel (HMS)
and the Retailer Scanner Data (RMS). The HMS data contains purchase histories of for a ro-
tating panel of households from 2004 to 2019. The RMS data contains anonymized purchases
of products aggregated to the store-week level throughout 2007 to 2017. We supplement the
NielsenIQ data with the Consumer Expenditure Survey public use micro data (CE PUMD)
to document descriptive evidence of differences in consumption spending across the entire
consumption basket. Lastly, we incorporate data from National Promotion Reports’ PRICE-
TRAK database (PromoData), which features data on wholesaler prices charged to retailers
for certain products from 2008-2013. While we discuss these data in turn, see B. J. Bron-
nenberg et al. (2015) and Allcott, Lockwood, and Taubinsky (2019) for further discussion of
the NielsenIQ data.
The entire HMS features data on the shopping trips and transactions of approximately 60k
10households per year. Households remain in the panel for on average 54 months, with ap-
proximately 200,000 distinct households rotating through the HMS in total. The data report
on purchases made by households on the 20 million shopping trips from 2004 to 2019 made
by the panelists. For each individual item purchase, we observe the transaction metadata
such as date, store/retailer-info, and panelist identifier, as well as granular data on product
and transaction details, including prices paid, amounts and units of quantities purchased,
deal/sale usage, and detailed nests of product identifiers.
Our primary uses for the HMS data are to document differences in the purchasing behav-
ior of men and women and understand how product markets differ for men and women.
To confidently assign product purchases to consumer gender demographics, we restrict our
consumer panel to single-individual households that log at least 12 shopping trips per year,
which eliminates approximately 75% of the panelists in the HMS. This leaves us with a panel
of 47,012 households which we use to study differences in consumer behavior. Summary
statistics for the sample can be found in 1. Our final sample is skewed women, with about
70% of our panelists identifying as a woman. In terms of balance, the men in our sample
tend to have higher income and be more educated, which we will control for in the analysis.
The second component of our analysis focuses on how the product market space varies by
gender. For this analysis, we restrict our data to products that we can confidently assign a
gender to. We describe our methodology in detail in Section 3.2. The NielsenIQ data covers
approximately 1.8 million products and we are able to confidently assign gender to 700,000
of them. However, these 700,000 products comprise 97% of the purchases made in our singles
panel.
There is considerable discussion on the representativeness of the HMS panel. B. J. Bronnen-
berg et al. (2015) summarize this discussion that argues in favor of the representativeness
of the panel of US consumers. While applying the included HMS projection weights render
the sample much more representative of the US, the raw using-sample departs significantly
from basic US demographics. Our sample skews significantly more female than male, by a
ratio of 3:1, and the in-sample median age of 53 is significantly older than the US median
11age of 38. The panelist’s income demographics appear slightly more representative, with
the median single-individual household earning approximately $37,000 USD per year and
the median household, unconditional, reporting approximately $55,000 USD.3 Nonetheless,
applying the projection weights yields demographics that much more closely align with those
of US consumers.
The RMS data contain product-store-week level prices and volumes of products purchased
Table 1: Demographics of HMS panelists sample of single-member households
Total Women Men Difference
Income 44687 39514 50682 -11167.86**
(37202.4) (34048.25) (39718.72) (340.2182)
Age 53.47 53.21 53.77 -.556**
(16.4528) (17.223) (15.5078) (.1522)
High school 0.602 0.637 0.562 .074**
(.4894) (.481) (.4961) (.0045)
College 0.238 0.206 0.275 -.069**
(.4258) (.4044) (.4464) (.0039)
Post-grad 0.120 0.115 0.127 -.012**
(.3255) (.3187) (.3332) (.003)
White 0.785 0.767 0.805 -.038**
(.4111) (.4228) (.3962) (.0038)
Black 0.133 0.157 0.106 .051**
(.3399) (.3636) (.308) (.0031)
Asian 0.0250 0.0220 0.0270 -.005**
(.155) (.1479) (.1627) (.0014)
Hispanic 0.0660 0.0670 0.0650 0.00200
(.2485) (.2503) (.2463) (.0023)
No. households 47012 33628 13384 20244
This table displays demographic data of men and women constituting single-member households as well as
their differences. These figures and their corresponding gender-differences were computed using the pro-
prietary analytic household weights included in the Nielsen Consumer Panel Survey. Dollar amounts are
expressed in USD 2016.
∗p < .05, ∗ ∗ p < .01
by consumers from 2004 to 2018. This dataset is not tied to the consumer identifiers; rather,
the strength of the RMS data is in its relative comprehensiveness of US sales. We use the
RMS data to model demand in select markets that have a high level of gendered products
(as identified in the HMS data). These markets include yogurt, health and protein bars,
3
These figures represent the midpoint of the discrete income buckets used for the household income field
12deodorant, disposable razors and shampoo. While the HMS tracks all retail purchases for a
household from any store, the RMS contains a select set of stores. For our analysis, we keep
only stores that are part of a larger retail chain rather than independent stores.
Both components of the NielsenIQ data feature a highly detailed product hierarchy clas-
sification that organizes all goods into smaller nests with increasing degrees of specificity.
Products in the NielsenIQ are identified with their Universal Product Code (UPC) which
corresponds to a unique barcode. All UPCs fit into one of ten departments (the broadest cat-
egory, e.g. “Health and Beauty” and ”Dry Grocery”). From here, products in a department
are allocated to Product Groups—of which there are 120 total—such as “Shaving Needs”.
Finally, UPCs in the same Product Group are assigned to Product Modules—the most gran-
ular grouping of multiple products—e.g. “Disposable Razors”. The Nielsen data identifies
over 1300 distinct product modules. Brand description represents an alternate grouping that
features the brand name for a given set of UPCs, not strictly contained in any single Product
Module or Group contained, such as “Venus”, for the brand of razors. We consider Product
Modules as constituting a self-contained goods market; for certain reduced-form analyses,
we further divide product modules into Module-Unit groups (modules composed of goods all
with the same counting units: e.g. the coffee product module contains bagged coffee mea-
sured in weight (ounces) and Keurig cup coffee measured as a count (number of K-cups).
The Consumer Expenditure Survey Public Use Microdata (CE PUMD) is publicly available
from the Bureau of Labor Statistics and provides information on a household’s expenditures
and income. The CE PUMD is comprised of a quarterly interview survey of 6,000 households
that tracks overall spending and large purchases and a diary survey of 3,000 households that
tracks all purchases over a two week period. We utilize only the quarterly interview surveys
to inform aggregate consumption basket price and composition differences. Similar to the
Nielsen HMS data, we restrict our analysis to individuals that live alone which allows us
to attribute spending to one gender. We use data from years 2010 to 2017 which comprise
67,950 person-quarter observations. Summary statistics are presented in Appendix Table
A.4. Similarly to our HMS single household panel, our CE PUMD single household panel
13shows that women tend to be older and poorer than the men in the sample, but otherwise
are roughly similar demographically. The CE PUMD interview survey contains quarterly
spending info for several categories; we focus on the eight categories that comprise the vast
majority of spending: food, housing, clothing, transportation, health, entertainment, per-
sonal care, and alcohol and cigarettes. Each category aggregates all of the spending made
by the individual in the quarter before their interview. Thus the food category contains all
spending related to food: groceries, restaurants, convenience stores, etc. The housing cate-
gory includes both rental and mortgage spending, health includes health insurance, payments
to health care providers and prescriptions, and personal care includes hygiene, well being and
beauty spending.
Lastly, the PriceTrak PromoData data allow us to validate retailer markups relative to whole-
saler price. A critical component of this work consists of assigning the source of the observed
differences in prices of goods consumed by women and by men to differences in marginal costs
and differences in markups. While this data does not feature information on production costs,
it does provide on the intermediary costs to retailers. The PriceTrak PromoData ultimately
serves to facilitate a cross-validation against our structural demand estimation that uses
solely price and consumption information from NielsenIQ. The PriceTrack data features re-
tailer cost-data of individual UPCs for a variety of time- and geographic-denominations from
2006 and 2012, with the geographic disaggregations covering 48 markets (coinciding with the
metropolitan areas around large US cities). The match rate of UPCs in the Promodata to the
NielsenIQ datasets is relatively low—with only about 18% of the 430,000 distinct UPCs in
the RMS data matching to PromoData. We combine the data from PriceTrak on wholesaler
prices with Nielsen data on post-deal consumer prices to compute retailer markups relative
to wholesaler prices following B. J. Bronnenberg et al. (2015).
4 Price Disparities Across the Consumption Bundle
We present evidence of gender differences in both consumer behavior and the product space.
We begin by examining overall differences in consumption basket composition, finding signif-
14icant differences in how men and women choose to allocate their income. We document that,
within grocery and big box retail purchases, women spend more than men—both overall and
per item—and that products primarily bought by women are priced higher than those bought
equally by men and women or primarily by men. These consumer behavior differences and
product space differences indicate that gender disparities in consumption are driven by both
demand and supply side forces. Women spend more per item, and there exists a larger prod-
uct space of goods marketed more exclusively toward women than toward men. In line with
these findings, we demonstrate the existence of a women’s price premium of approximately
4% on average.
4.1 Consumer Behavior by Gender
First, we document that women’s consumption bundles are different from those of men in
terms of composition. Using the CE PUMD we find that women and men do not have sig-
nificant differences in total yearly spending, but how they choose to allocate their spending
highlights important differences in preferences across all types of spending. Figure 1 plots
women’s yearly spending as a percentage of men’s. Each bar plots the coefficient from a
regression of log spending for a category on an indicator for the individual being a woman
controlling for age, income and race. Women spend significantly more of their income on
housing, clothing, health and personal care, while men spend relatively more on food, alcohol
and cigarettes, and transportation. These findings roughly correspond with markets that
are often discussed in discourse on the pink tax and gendered marketing more broadly. The
focus of this paper is on differences in men and women’s behavior and product space for retail
markets like grocery and big box stores. These purchases largely fall under the categories
of food, alcohol and cigarettes, and personal care but they do not map perfectly. A key
descriptive result of our paper is that women spend more on retail purchases than do men; in
the context of figure 1 this would imply that women spend more on food as groceries while
men spend more on food out. Similar overall levels of spending with differing allocation
patterns highlights the important role that preferences, substitution patterns and societal
expectations play in evaluating the pink tax.
15Figure 1: Women’s yearly consumption spending relative to men’s
2 1.80
Proportion of Men's Activity
1.5
1.32
1.11 1.11
0.99
0.91 0.94
1
0.67
.5 0
od
s
ng
g
n
lth
t
e
en
te
in
tio
ar
ea
Fo
si
et
th
nm
C
rta
ou
ar
lo
H
al
po
i
H
C
ta
ig
on
C
r
ns
te
rs
nd
a
En
Pe
Tr
la
ho
co
Al
Note: This figure plots the coefficients estimated from a regression of log expenditure on an indicator for the individual being
a woman and demographic controls: log yit = β · 1{womani = 1} + ΓXi,t + εi,t , for spending categories food, alcohol and
cigarettes, housing, clothing, transportation, health entertainment and personal care using the CE PUMD from 2010 to 2017.
1{f emalei = 1} is an indicator for whether the individual is a woman, and Xi,t is a vector of time- and time-id-varying controls
including income, age, race and education. Standard errors are clustered at the individual-level.
While figure 1 speaks to full consumption basket differences between men and women, we
now turn our focus to retail spending consumption baskets and how they vary by gender.
We find that women’s retail consumption baskets are larger, more expensive, and filled with
a greater number of unique UPCs. Figure 2 plots levels of female activity as a proportion
of male activity for annual spending, unique product purchases, and overall product pur-
chases. We find that women’s yearly spending is greater than that of men by about 6%,
their product diversity is greater than men’s by about 27% and their consumption baskets
are larger than men’s in terms of items purchased by about 9%. This pattern is primar-
ily driven by differences in behavior in consumption of Health and Beauty products, where
women spend 51% more than men, consume 53% more unique products, and consume 49%
more items. However, we observe similar results for all products after excluding Health and
16Beauty; such spending categories include are food grocery products, household products and
alcohol. Among these products women spend about 2% more, have 25% greater product
diversity and 7% more items than men.
Figure 2: Women’s yearly retail consumption spending relative to men’s
1.51 1.53
1.5
1.49
1.27 1.25
Proportion of Men's Activity
1.09 1.07
1.06
1.02
.5 0 1
Spending Unique Products Total Items
All Health & Beauty Not Health & Beauty
Note: This figure plots the coefficients estimated from a regression of log expenditure on an indicator for the individual being
a woman and demographic controls: log yit = α + β · 1{womani = 1} + ΓXi,t + εi,t , for dependent variables including yearly
spending, unique products purchased, and total items purchased. 1{womani = 1} is an indicator for whether the individual is a
woman, and Xi,t is a vector of time- and time-id-varying controls including income, county, age, race and education. Standard
errors are clustered at the individual-level.
Figure 2 compares men and women that otherwise look demographically similar in terms
of location, age, race, income and education. Our panel of singles is weighted to be rep-
resentative of all single men and women in the United States, and thus we can also make
comparisons of how men and women’s spending differs in the aggregate, by location, etc.
Table 2 reports yearly spending differences between men and women subsequently adding in
these demographic controls. Column (1) reports aggregate differences in spending between
men and women, including only controls for year. We find small differences in yearly spend-
ing without demographic controls of about 1.6% higher yearly spending by women. Columns
172, 3, 4, 5, and 6 add in fixed effects for county of residence, income, age, race and education
respectively. Column (2) compares yearly spending between men and women that live in
the same county, finding 2.5% higher spending by women. We can interpret the increase in
the magnitude of the coefficient across columns (1) and (2) as the contribution of geographic
sorting of single men and single women to overall spending differences and is consistent with
single women more often living in lower cost areas than do men. This interpretation continues
as we move to columns (3) and (4) which add in controls for income and age which raise the
spending gap to 4.4% and 6.2% respectively. Because single women tend to skew lower in-
come and older in age than single men, we can see the attenuating effect that lower spending
among older and poorer women has in the aggregate. Columns (4) and (5) add in controls
for race and education; while racial composition differences between single men and single
women do contribute to the spending gap somewhat, the magnitude of the change is much
smaller than the contribution from geography, age and income. Controlling for education
has no contribution that cannot be accounted for by geography or the other demographic
variables. While yearly spending differences vary significantly across different comparisons of
interest, the same analysis on the number of unique products consumed or total number of
items purchased in a year shows little variation (see appendix X). These findings suggest that
while there are many factors that contribute to yearly spending differences between men and
women, gender differences in consumption basket size and composition are fairly constant.
Figure 2 and 2 document that women’s consumption bundles differ from those of men in
important ways, but does not fully inform the way through which a pink tax may take form.
Aggregate spending differences can arise from differences in prices paid for similar goods
or from differences in quantities purchased. As a clarifying example, consider consumption
habits for shampoo. Women, on average, have longer hair than men which may lead them
to buy more bottles of shampoo over the course of a year, we refer to this as driving up
total spending on the extensive margin, that is, buying more product. It is also possible
that women have preferences for higher priced shampoos, we refer to this as the intensive
margin, where women are paying higher per unit prices. Figure 2 indicates that the “exten-
sive” margin is an important contributor to overall differences in spending. While total items
18Table 2: Yearly spending differences between men and women
(1) (2) (3) (4) (5) (6)
∗∗
Women 0.0162 0.0248∗∗∗ 0.0444∗∗∗ 0.0616∗∗∗ 0.0677∗∗∗ 0.0677∗∗∗
(0.0080) (0.0077) (0.0076) (0.0076) (0.0075) (0.0075)
Observations 216890 216743 216743 216742 216742 216742
Adjusted R2 0.018 0.097 0.105 0.125 0.133 0.133
Year FE Yes Yes Yes Yes Yes Yes
County FE No Yes Yes Yes Yes Yes
Income FE No No Yes Yes Yes Yes
Age FE No No No Yes Yes Yes
Race FE No No No No Yes Yes
Education FE No No No No No Yes
Individual level clustered standard errors in parentheses
∗
p < .10, ∗∗ p < .05, ∗∗∗ p < .01
Note: This table presents estimates of the percent difference in yearly spending between men and women using the following
regression: log yit = φt +β · 1{womani = 1}+ΓXi +εi,t , where it is yearly spending. 1{womani = 1} is an indicator for whether
the individual is a woman, φt is a time fixed effect and Xi is a vector of demographic controls including income, county, age,
race and education which we add in sequentially. Standard errors are clustered at the individual-level. Column 1 can be thought
of as a raw gap between single men and single women, each subsequent column demonstrates the contribution of controlling for
an additional demographic factor.
purchased captures the differences both in the intensity and variety of products purchased,
information on unique products captures only this latter element, and could be driven by
both greater taste for variety by women within shared-gender product spaces as well as a
greater volume of products typically intended for exclusive consumption by women (e.g. fem-
inine hygiene products, medication and beauty products).
Popular discussion of the pink tax is often focused on differences in prices paid between
men and women, the intensive margin contribution to the overall spending gap. We compare
per unit prices paid by men and women for products in the same market with the following
specification:
log(Pijt ) = φt(j) + β1w(i) + γXi + ijt
Where i denotes the individual, j denotes the product purchased and t denotes the market.
Table 3 presents the results. Column (1) regresses log unit UPC price on a woman indicator
and includes fixed effects for the interaction of product module, units the good is sold in and
the year of purchase. Similar to Table 2, we can think of the 2.3% result as the raw differ-
19ence in prices paid between single men and women, not accounting for other demographic
factors or location and retail chain sorting. Column (2) runs the same specification but
adds in controls for age, income, and race. The large increase in the coefficient, from 2.3% to
4.67% highlights the important role of demographic differences between single men and single
women because older and lower income people tend to buy lower priced products. Columns
(3) and (4) subsequently add in county and retailer fixed effects. We can think of Column
(3) as the contribution of women sorting into more or less expensive locations, because the
coefficient change is small, the contribution is minimal. Similarly, column (4) can be thought
of as the contribution of sorting into more or less expensive retail chains, i.e. Whole Foods
vs. Walmart. Controlling for the retail chain lowers our price premium estimate to 4.19%,
suggesting that retail chain sorting plays a small but significant role. Finally, in Column (5)
we add in fixed effects for month rather than year. The results indicate that women spend
more than 4.02% more than do men per unit of goods in the same product market, bought
in the same retail chain, county, and month. We consider this our preferred specification
because it attempts to control, as much as possible, for all potential differences that could
arise between the two groups other than gender.
We refer to this 4% finding as our observed pink tax on the intensive margin. This price
premium could be driven by many different factors. First, it can be driven by women buying
products that are made specifically for and marketed to women, this would be in line with
how the pink tax is traditionally thought about. Alternatively, it could be driven by differ-
ences in preferences between men and women for products that are otherwise ungendered.
That is, if women happen to like organic products or name brand products more than men
then we would likely observe that women pay higher per unit prices than do men. Once
we understand which types of products are contributing to our observed pink tax, we want
to know whether these price premiums are being driven by markups or marginal costs. The
underlying implication of popular discourse on the pink tax is that it is a price discrimination
story: two products differ only in their color but the pink product is priced higher, because
their costs of production must be the same the women’s product faces a higher markup. This
price discrimination story would require that women either consume products less elastically
20or the competitive structure of the market is such that the products women buy face higher
markups. The alternative explanation is that the products that women buy have higher
marginal costs of production. This would be consistent with women having preferences for
higher “quality” goods.4 The rest of the paper strives to understand what generates our
4% pink tax by analyzing how product markets vary for men and women and estimating
differences in demand between men and women.
Table 4 estimates the same equation as Table 3 while including product level fixed effects
Table 3: Unit prices in same product module
(1) (2) (3) (4) (5)
Woman 0.0230∗∗∗ 0.0467∗∗∗ 0.0512∗∗∗ 0.0419∗∗∗ 0.0402∗∗∗
(0.0035) (0.0033) (0.0028) (0.0020) (0.0018)
Observations 153,333,409 153,333,409 150,059,493 143,532,160 139,739,839
Adjusted R2 0.829 0.831 0.868 0.889 0.877
ModuleXUnits FE Yes Yes Yes Yes Yes
Year FE Yes Yes Yes Yes No
County FE No No Yes Yes Yes
Retailer FE No No No Yes Yes
Month FE No No No No Yes
Demographic FE No Yes Yes Yes Yes
Individual level clustered standard errors in parentheses
∗
p < .10, ∗∗ p < .05, ∗∗∗ p < .01
Note: This table presents estimates from the regression: log(Pijt ) = φt(j) + β1w(i) + γXi + ijt where Pijt is the per-unit price
of a UPC. 1{womani = 1} is an indicator for whether the individual is a woman, φt is a market-time fixed effect and Xi is a
vector of demographic controls including income, county, age, race and education which we add in sequentially. Standard errors
are clustered at the individual-level. Column 1 can be thought of as a raw gap between single men and single women, each
subsequent column demonstrates the contribution of controlling for an additional market or demographic factor.
instead of module level fixed effects:
log(Pijt ) = φjt + β1w(i) + γXi + ijt
The interpretation of the coefficient becomes the difference in prices paid between men and
women for the same exact product. Differences in prices paid for the same good can be
attributed to differences in price shopping behavior, like coupon usage and sale shopping,
4
We cannot directly attribute higher marginal costs to higher quality as quality is likely not fully innate
but perceived by the individual.
21consistent with being a more elastic consumer. We sequentially add in fixed effects in the
same manner as Table 3, so the coefficients can be interpreted as a raw difference between
men and women in column (1) and then iteratively making comparisons between demograph-
ics, location, retail chain and month. Just like Table 3 we find that demographic differences
and differential sorting into retail chains and locations contribute to the price shopping gap.
While we find that women, on average, buy more expensive products than do men, we find
that they spend 0.8% less than men on the same product. Column (5) captures differences
in prices paid for the same product by people that differ only in gender over a month, which
we attribute to differences in price shopping behavior. Combining this with our result from
Table 3 suggests that women are buying higher priced goods while also exhibiting behaviors
associated with being more elastic consumers. Hendel and Nevo (2013) study promotional
sales as a form of intertemporal price discrimination, our results would indicate that women
are likely to comprise a larger share of the consumer base that benefits from this type of price
discrimination.
Table 4: Unit prices for the same product
(1) (2) (3) (4) (5)
Women -0.0089∗∗∗ -0.0055∗∗∗ -0.0060∗∗∗ -0.0075∗∗∗ -0.0080∗∗∗
(0.0017) (0.0017) (0.0015) (0.0010) (0.0010)
Observations 151,188,750 151,191,277 139,671,522 138,165,657 135,154,990
Adjusted R2 0.952 0.840 0.860 0.878 0.879
UPC FE Yes Yes Yes Yes Yes
Year FE Yes Yes Yes Yes No
County FE No No Yes Yes Yes
Retailer FE No No No Yes Yes
Month FE No No No No Yes
Demographic FE No Yes Yes Yes Yes
Individual level clustered standard errors in parentheses
∗
p < .10, ∗∗ p < .05, ∗∗∗ p < .01
This table estimates reduced forms specified as log(Pijt ) = φjt + β1w(i) + γXi + ijt where Pijt is the per-unit price of a
UPC. 1{womani = 1} is an indicator for whether the individual is a woman, φt is a UPC-market-time fixed effect and Xi is a
vector of demographic controls including income, county, age, race and education which we add in sequentially. Standard errors
are clustered at the individual-level. Column 1 can be thought of as a raw gap between single men and single women, each
subsequent column demonstrates the contribution of controlling for an additional market or demographic factor.
Table 5 estimates the preferred specifications from tables 3 and 4, stratifying by department.
22Table 5: Prices paid across departments
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
H&B Dry Groc. Frozen Dairy Deli Pack. Meat Produce Non-food Groc. Alcohol Gen. Merch.
Panel A: Per unit prices within product module
Female 0.0534∗∗∗ 0.0546∗∗∗ 0.0489∗∗∗ 0.0374∗∗∗ 0.0348∗∗∗ 0.0485∗∗∗ 0.0182∗∗∗ 0.0348∗∗∗ -0.1450∗∗∗ -0.0484∗∗∗
(0.0029) (0.0019) (0.0025) (0.0022) (0.0063) (0.0031) (0.0037) (0.0021) (0.0204) (0.0039)
Observations 10504032 55328879 14261546 16609254 5270903 4094787 10769045 12467554 1991861 6719882
Adjusted R2 0.836 0.860 0.903 0.934 0.900 0.798 0.779 0.870 0.637 0.780
ModXUnitXRetXLocXMonth FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Demographic FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Panel B: Per unit price for same UPC
Woman -0.0212∗∗∗ -0.0034∗∗∗ -0.0050∗∗∗ -0.0012 -0.0258∗∗∗ -0.0070∗∗∗ -0.0160∗∗∗ -0.0133∗∗∗ -0.0031 -0.0016
(0.0022) (0.0009) (0.0014) (0.0010) (0.0055) (0.0015) (0.0030) (0.0010) (0.0019) (0.0049)
Observations 9668836 61456951 12812841 15406630 5381334 3922662 10758430 10671760 1891549 6259655
23
2
Adjusted R 0.817 0.891 0.842 0.875 0.637 0.813 0.654 0.933 0.930 0.854
UPCXRetXLocXYear FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Demographic FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Individual level clustered standard errors in parentheses
∗
p < .10, ∗∗ p < .05, ∗∗∗ p < .01
This table estimates log(Pijt ) = φt + β1w(i) + γXi + ijt , stratifying by department across columns. Pijt is the per-unit price of a UPC. 1{womani = 1} is an indicator for
whether the individual is a woman and and Xi is a vector of demographic controls including income, county, age, race. In panel A, φt is a vector of fixed effects for the interaction
of product module, units, retailer chain, county, and half-year. In Panel B φt is a vector of fixed effects for the interaction of product (UPC), retailer chain, county, and half-year.You can also read
